Detection of burned areas in Africa using a multitemporal multithreshold analysis of NOAA AVHRR GAC data

A methodology to detect vegetation burned areas is presented together with the results obtained for the African continent between November 1990 and October 1991. NOAA-AVHRR-GAC-5 Km images were used in this study. The spectral bands and indices used were land surface temperature (Ts) and the Global Environment Monitoring Index (GEMI). The time series was composited in weekly images using the minimum value composite of albedo (MiVCA). After analysis of the weekly profiles on the main vegetation types that are affected by burning, a multitemporal multithreshold technique to detect burned pixels was developed [Burned Area Algorithm (BAA)]. This technique was based on the increase of Ts and a decrease in the GEMI after a fire occurrence. The results showed good agreement at the continental scale with the temporal and spatial patterns of active fires from the IGBP-DIS Global Fire Product. Comparison with a Landsat TM image classification showed good performance of the algorithm.